Benefits of in-vehicle consolidation in less than truckload freight transportation operations Rodrigo Mesa-Arango Satish Ukkusuri 20th International Symposium of Transportation and Traffic Theory Noorwijk, Netherlands July 2013 Outline 1. 2. 3. 4. 5. 6. Introduction Problem Methodology Numerical Results Conclusion Questions/Comments 2 / 28 1. Introduction • Trucking: Important economic sector (1) – US GDP: • For hire transportation: – Trucking: – Air: – Rail: $ 14,499 billion dollars $ 403 billion dollars $ 116 billion dollars $ 63 billion dollars $ 15 billion dollars • Externalities - Emissions - Safety - Congestion - Asset deterioration • Mitigation: Increasing vehicle utilization(2)(3)(4)(5) (1) U.S. Department of Transportation (2012). National transportation statistics (2) Sathaye, et al, The Environmental Impacts of Logistics Systems and Options for Mitigation, 2006 (3) Organisation for Economic Co-Operation and Development. Delivering the Goods-21st Century Challenges to Urban Goods Transport. 2003. (4) European Commission, Directorate-General for Energy and Transport. Urban Freight Transport and Logistics. European Communities. 2006. (5) Transport for London. London Freight Plan – Sustainable Freight Distribution: A Plan for London. 2007. 3 / 28 1. Introduction • Economic mechanism attractive for consolidation? • Combinatorial Auctions – Successful implementations (6)(7)(8)(9)(10): - Home Depot Inc. - Staples Inc. - Wal-Mart Stores Inc. - Reynolds Metal Company - K-Mart Corporation - Ford Motor company - The Limited - Compaq Computer Corporation - Sears Logistics Services (6) Elmaghraby, and Keskinocak. Combinatorial Auctions in Procurement. 2002. (7) De Vries, and Vohra. Combinatorial Auctions: A Survey. 2003. (8) Moore, et al. The Indispensable Role of Management Science in Centralizing Freight Operations at Reynolds Metals Company. 1991 (9) Porter, et al. The First Use of a Combined-Value Auction for Transportation Services. 2002. (10) Sheffi, Y. Combinatorial Auctions in the Procurement of Transportation Services. 2004. 4 / 28 1. Introduction • Combinatorial Auctions in Freight Transportation Shipper Winner Determination Problem (11) (12)(13)(14) (11) Caplice and Sheffi. Combinatorial Auctions for Truckload Transportation. 2006. (12) Sandholm. Algorithm for Optimal Winner Determination in Combinatorial Auctions. 2002 (13) Abrache, et al. Combinatorial auctions. Annals of Operations Research. 2007 (14) Ma, et al. A Stochastic Programming Winner Determination Model for Truckload Procurement Under Shipment Uncertainty. 2010 5 / 28 1. Introduction • Bidding advisory models − Truckload (TL) operations(15)(16)(17)(18)(19) • Direct movements • Economies of scope(20)(21)(22)(23) − Less-Than-Truckload (LTL) operations? • Consolidated movements • Economies of scope, scale, density(20)(21)(22)(23) (15) Song, and Regan. Combinatorial Auctions for Transportation Service Procurement, The Carrier Perspective. 2003, (16) Song, and Regan. Approximation Algorithms for the Bid Construction Problem in Combinatorial Auctions for the Procurement of Freight Transportation Contracts. 2005, (17) Wang, and Xia. Combinatorial Bid Generation Problem for Transportation Service Procurement. 2005 (18) Lee, et al. A Carrier’s Optimal Bid Generation Problem in Combinatorial Auctions for Transportation Procurement. 2007 (19) Chang. Decision Support for Truckload Carriers in One-Shot Combinatorial Auctions. 2009 (20) Caplice, and Sheffi. Combinatorial Auctions for Truckload Transportation. 2006 (21) Caplice. An Optimization Based Bidding Process: A New Framework for Shipper-Carrier Relationship. 1996 (22) Jara-Diaz. Transportation Cost Functions: A Multiproducts Approach. 1981 (23) Jara-Diaz. Freight Transportation Multioutput Analysis. 1983 6 / 28 1. Introduction • Routes, costs and prices… 7 / 28 1. Introduction • Economies of scope [TL] c12 (i ) 1 2 h c24 c41 4 (i ) c12 c34 (ii ) c23 1 3 4 (ii ) 2 k c41 c12 c24 c34 1 (iii) c23 h k 3 c41 2 c24 4 c34 c23 h 3 (iii) 8 / 28 1. Introduction • Economies of consolidation (scale and density) [LTL] c12 (i ) 1 4 (ii ) 2 h c24 c41 (i ) c12 c34 c23 1 4 3 (ii ) 2 k c41 c12 c24 c34 1 (iii) c23 h k 3 c41 2 c24 4 c34 c23 h 3 (iii) 9 / 28 1. Introduction • This research – Show Benefits for carries • In-vehicle consolidation • Bidding construction – Freight Transportation combinatorial auctions – Use • Multi-commodity one-to-one pick up and delivery vehicle routing problem (m-PDVRP) to find optimal LTL bundles. – Compare against optimal bundles obtained for TL carriers 10 / 28 2. Problem • MIP Formulation for m-PDVRP (1/2) min vV (i , j )A x vV jN x jN v 0j x jN v ji 1; i N ' v ij Objective fun: Minimize total traversing cost Each node visited once 1; v V All vehicles are used xijv ; i N ; v V Vehicle flow conservation x iM jM cij xijv jN v ij M 1; M N '; v V xijv {1,0}; i, j N; v V Sub-tour elimination Binary variables … 11 / 28 2. Problem • MIP Formulation for m-PDVRP (2/2) min vV (i , j )A cij xijv Objective fun: Minimize total traversing cost … l vV jN ri ,v ji pri ; r N ; i N ' Demand Satisfaction constraint (Deliveries) is ,v ij pis ; s N ; i N ' Demand Satisfaction constraint (Pickups) l vV jN l jN lijrs,v ; i N '; r, s N \ i; v V rs ,v ji l rN sN Payload flow conservation jN rs ,v ij xijv Q; i, j N '; v V Loads only on traversed links without exceeding vehicle capacity l1rsi ,v 0 and lirs1 ,v 0; i N '; r, s N \ i; v V Vehicles leave the depot empty and return empty lijrs,v 0; i N , j N , r N , s N , v V Non-negative continuous variables 12 / 28 3. Methodology • Branch-and-price(24)(25) – Branch-and-bound – Dantzig-Wolfe and Column generation • Master Problem • Sub - problem (24) Barnhart, et al. 1998. Branch-and-price: Column generation for solving huge integer programs. (25) Desaulniers, et al. 1998. A unified framework for deterministic time constrained vehicle routing and crew scheduling problems. 13 / 28 3.1 Branch-and-bound • Branch-and-Bound – Solve linear relaxation of IP – Terminate (fathom) a node • Infeasibility \ Bound \ Solution – Branch – Stop when all nodes are terminated IP Linear Relaxation Branch-and-Price – Dantzig Wolfe Decomposition – Column Generation LP0 LP1 LP2 14 / 28 3.2. Dantzig Wolfe dec. + col. gen. • MIP has special structure appropriate for decomposition – Master Problem (MP) • • • • Linear Program Controls column generation process Requests columns from the Sub problem Integer variables are represented as convex combination of the columns generated by the Sub problem – Sub Problem • Integer program • Generates columns – Set of integer variables with common structure 15 / 28 3.2. Dantzig Wolfe dec. + col. gen. x vV jN x jN v 0j x jN v ji Each node visited once 1; v V All vehicles are used xijv ; i N ; v V Vehicle flow conservation x iM jM 1; i N ' v ij jN v ij M 1; M N '; v V xijv {1,0}; i, j N; v V Sub-tour elimination Binary variables … Convex combination xijv xijv (t ) t VRP deployment (t) tT t tT 1 t 0 16 / 28 3.2. Dantzig Wolfe dec. + col. gen. • Examples of deployments of trucks |V| = 1 0 |V| = 2 1 1 0 1 t=1 3 2 3 2 0 1 0 1 3 2 t=2 t=2 3 1 t=1 t=1 0 0 |V| = 3 2 3 2 3 2 0 1 0 1 t=3 t=3 3 2 3 2 17 / 28 3.2. Dantzig Wolfe dec. + col. gen. • Master problem (MP): Generates t as needed min c Objective fun: Minimize total traversing cost l jN tT jN is ij l jN pri ; r N ; i N ' ri ji l rs ji (MP) Demand Satisfaction constraint (Deliveries) pis ; s N ; i N ' Demand Satisfaction constraint (Pickups) lijrs ; i N '; r, s N \ i Payload flow conservation l rN sN t t rs ij jN Q aijt t ; i, j N ' tT ( ij ) Loads only on traversed links without exceeding vehicle capacity l 0 and lirs1 0; i N '; r, s N \ i Vehicles leave the depot empty and return empty lijrs 0; i N , j N , r N , s N Non-negative continuous variables l1rsi 0 and lirs1 0; i N '; r, s N \ i Vehicles leave the depot empty and return empty rs 1i jN t 1 t 0 t T ( 0 ) Convexity Constraint Non-negativity 18 / 28 3.2. Dantzig Wolfe dec. + col. gen. • Each Solution generates a column t, {x0j0,…,xi0v}, that is associated with a variable t in the MP c min * c vV ( i , j )A x vV jN x jN v 1j x jN v ji ij xijv 0 1; i N ' v ij Objective fun: Minimize reduced cost (Sub-P) Each node visited once 1; v V All vehicles are used xijv ; i N ; v V Vehicle flow conservation x iM jM ij jN v ij M 1; M N '; v V xijv {1,0}; i, j N; v V Sub-tour elimination Binary variables 19 / 28 3.2. Branch-and-price Root B&B Node (Active) Solve MP Update arcs and costs Set Sub-P costs Add new column to pool Solve Sub-P No Column Generation Yes Select B&B node and set as inactive Terminate node by infeasibility Terminate node by bound Terminate node by solution Active nodes? Set node as inactive No Update incumbent solution Branch B&B node (active) Column Generation B&B node (active) Column Generation Stop 20 / 28 3.3. Acceleration strategies • Originally depth-first search – Finding initial incumbent solution (upper bound): Time consuming • Strategy 1: Fast initial upper bound • Strategy 2: Continuous increment to lower bound – Strategy 1 replace Step 3 as follows • Find branch-and-bound node with current lowest solution and fathom it, repeat 21 / 28 4. Numerical Results • Implementation – Java • Branch-and-Bound • Interactions in Column Generation – Set Sub-P – Update MP • Network Management – Information/Updates: Nodes, Links, Tours – ILOG CPLEX • MP LP Solution • Sub-P IP Solution 22 / 28 4. Numerical Results Scenario 1 Scenario 2 1 1 10 2 20 10 5 2 20 20 0 0 4 1 7 10 10 5 4 3 Scenario 3 2 20 20 0 6 4 8 3 6 cij 0 1 2 3 4 5 6 7 8 3 0 1 2 3 4 5 6 7 8 99.0 3.0 7.0 5.0 1.0 1.0 7.0 5.0 3.0 3.0 99.0 3.0 7.0 5.0 1.0 5.0 1.0 7.0 7.0 3.0 99.0 3.0 7.0 5.0 1.0 1.0 5.0 5.0 7.0 3.0 99.0 3.0 7.0 1.0 5.0 1.0 1.0 5.0 7.0 3.0 99.0 3.0 5.0 7.0 1.0 1.0 1.0 5.0 7.0 3.0 99.0 7.0 3.0 5.0 7.0 5.0 1.0 1.0 5.0 7.0 99.0 3.0 3.0 5.0 1.0 1.0 5.0 7.0 3.0 3.0 99.0 7.0 3.0 7.0 5.0 1.0 1.0 5.0 3.0 7.0 99.0 23 / 28 4. Numerical Results (LTL) Min. Deployment Cost 5 5 5 5 5 5 7 7 7 7 7 7 7 7 7 9 9 9 9 9 9 9 9 9 9 9 9 1 1 1 2 2 2 1 1 1 2 2 2 3 3 3 1 1 1 2 2 2 3 3 3 4 4 4 50 40 20 50 40 20 50 40 20 50 40 20 50 40 20 50 40 20 50 40 20 50 40 20 50 40 20 13 13 21 30 30 30 11 15 31 28 28 32 45 45 45 13 13 31 26 26 30 43 43 43 60 60 60 0-1-2-3-4-0 0-1-2-3-4-0 0-1-3-2-4-0 0-1-3-0-2-4-0 0-1-3-0-2-4-0 0-1-3-0-2-4-0 0-5-1-2-6-3-4-0 0-5-1-6-2-3-4-0 0-5-6-1-3-2-4-0 0-2-4-0-5-1-6-3-0 0-2-4-0-5-1-6-3-0 0-1-3-0-5-6-2-4-0 0-1-3-0-2-4-0-5-6-0 0-1-3-0-2-4-0-5-6-0 0-1-3-0-2-4-0-5-6-0 0-5-1-7-6-2-3-8-4-0 0-5-1-7-6-2-3-8-4-0 0-5-6-7-1-8-3-2-4-0 0-2-4-0-5-1-7-6-3-8-0 0-2-4-0-5-1-7-6-3-8-0 0-1-7-3-8-0-5-6-2-4-0 0-2-4-0-5-1-6-3-0-7-8-0 0-2-4-0-5-1-6-3-0-7-8-0 0-1-7-3-8-0-2-4-0-5-6-0 0-1-3-0-2-4-0-5-6-0-7-8-0 0-1-3-0-2-4-0-5-6-0-7-8-0 0-1-3-0-2-4-0-5-6-0-7-8-0 Time (sec) Bundles Deep-first search {(1,3),(2,4)} 0.203 {(1,3),(2,4)} 0.188 {(1,3),(2,4)} 1.640 {(1,3)},{(2,4)} 1.063 {(1,3)},{(2,4)} 1.094 {(1,3)},{(2,4)} 0.891 {(1,3),(2,4),(5,6)} 0.359 {(1,3),(2,4),(5,6)} 2.609 {(1,3),(2,4),(5,6)} 1.937 {(2,4)},{(1,3),(5,6)} 23.124 {(2,4)},{(1,3),(5,6)} 16.734 {(1,3)},{(2,4),(5,6)} 7.390 {(1,3)},{(2,4)},{(5,6)} 15.344 {(1,3)},{(2,4)},{(5,6)} 14.203 {(1,3)},{(2,4)},{(5,6)} 5.484 {(1,3),(2,4),(5,6),(7,8)} 19.203 {(1,3),(2,4),(5,6),(7,8)} 7.094 {(1,3),(2,4),(5,6),(7,8)} 53.312 {(2,4)},{(1,3),(5,6),(7,8)} 574.012 {(2,4)},{(1,3),(5,6),(7,8)} 383.779 {(1,3),(7,8)},{(5,6),(2,4)} 5.812 {(2,4)},{(1,3),(5,6)},{(7,8)} 2148.677 {(2,4)},{(1,3),(5,6)},{(7,8)} 1267.945 {(2,4)},{(1,3),(7,8)},{(5,6)} 205.436 {(1,3)},{(2,4)},{(5,6)},{(7,8)} 692.870 {(1,3)},{(2,4)},{(5,6)},{(7,8)} 503.496 {(1,3)},{(2,4)},{(5,6)},{(7,8)} 260.092 Strategy 1 Strategy 2 0.171 0.188 0.734 1.125 1.078 0.672 0.234 8.062 4.688 13.469 16.109 7.109 3.985 13.406 3.719 37.265 53.656 103.359 219.905 214.186 36.406 254.654 270.67 91.984 91.375 101.702 59.71 0.313 0.406 0.531 0.265 0.203 0.235 0.281 1.718 5.859 4.390 4.781 6.000 1.812 1.813 1.125 12.188 10.531 55.406 113.172 174.281 130.657 397.782 413.016 138.625 58.084 77.581 39.563 Gap (%) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 24 / 28 4. Numerical Results • TL + Scenario 3 Min. Cost Deployment 9 9 9 9 1 2 3 4 43 42 47 60 0-1-3-2-4-5-6-7-8-0 0-1-3-2-4-0-5-6-7-8-0 0-1-3-0-5-6-2-4-0-7-8-0 0-1-3-0-2-4-0-5-6-0-7-8-0 Bundles Time (sec) Gap (%) {(1,3),(2,4),(5,6),(7,8)} {(1,3),(2,4)},{(5,6),(7,8)} {(1,3)},{(2,4),(5,6)},{(7,8)} {(1,3)},{(2,4)},{(5,6)},{(7,8)} 14.000 25.266 113.172 78.188 0.00 0.00 0.00 0.00 • Comparison LTL operation No. Bundle Total lanes Deployment cost LTL {(1,3),(5,6),(7,8)} 3 0-5-1-7-6-3-8-0 11.00 LTL {(1,3),(5,6)} 2 0-5-1-6-3-0 13.00 TL {(1,3),(2,4)} 2 0-1-2-3-4-0 13.00 TL {(5,6),(7,8)} 2 0-5-7-6-8-0 13.00 TL {(5,6),(2,4)} 2 0-5-2-6-4-0 13.00 TL / LTL {(1,3),(2,4),(5,6),(7,8)} 4 0-5-1-7-6-2-3-8-4-0 13.00 TL / LTL {(1,3)} 1 0-1-3-0 15.00 TL / LTL {(2,4)} 1 0-2-4-0 15.00 TL / LTL {(5,6)} 1 0-5-6-0 15.00 TL / LTL {(7,8)} 1 0-7-8-0 15.00 Opt. for TL operation Cost per Total Deployment lane cost 3.67 0-5-6-1-3-7-8-0 35.00 6.50 0-5-6-1-3-0 25.00 6.50 0-1-3-2-4-0 21.00 6.50 0-5-6-7-8-0 21.00 6.50 0-5-6-2-4-0 17.00 3.25 0-1-3-2-4-5-6-7-8-0 43.00 15.00 0-1-3-0 15.00 15.00 0-2-4-0 15.00 15.00 0-5-6-0 15.00 15.00 0-7-8-0 15.00 Cost per lane 11.67 12.50 10.50 10.50 8.50 10.75 15.00 15.00 15.00 15.00 LTL min margin 24.01 12 8 8 4 30 0 0 0 0 25 / 28 5. Conclusion • Research shows benefits of considering in-vehicle consolidation (LTL) in the construction of bids • Numerical results show that consolidated bids (LTL) dominate non-consolidated ones (TL) • LTL carriers can submit bids with prices that are less than or equal to the costs of TL carriers • Savings increase as the capacity of trucks increases • Low transportation costs potentially reduce shipper procurement cost • In-vehicle consolidation (as defined in this research) integrates the flexibility of TL (economies of scope) to the economies of scales/density of LTL 26 / 28 5. Conclusion • Future research – Understanding the tradeoff between low price and delivery times (as well as other attributes of the carrier) for shippers • Econometric techniques • Segmented pricing policies – Acceleration of the solution methodology • Parallel computing • Hybrid-metaheuristics – Consideration of stochastic demand – Development of a robust biding advisory model that incorporates these features. – Analysis of positive/negative externalities associated to large trucks at a macroscopic level • Thank you! 27 / 28 6. Questions - Comments 28 / 28